对IP核进行扫描链平衡设计是缩短SoC测试时间的根本办法之一。扫描链平衡方法主要有BFD、MVA、MVAL等,但这些方法都是对扫描链的一次性分配,随着分配的深入容易陷入分配死区。同时在现有的方法中,MVA法在分配时利用扫描链平均值对结果进行预估和控制,因而能得到相对更好的结果,当遇到内部扫描链离散程度较大时却差强人意。针对上述问题作出改进:将扫描链切割成更易于分配的部分,利用每部分的平均值实现内部扫描链的再次优化重构,从而更易实现平衡分配。基于此提出了一种自动分段离均差的扫描链平衡设计方法,该方法主要分4步进行,首先根据扫描链长度的断层进行自动分段,其次找到每段扫描链中最接近平均值的扫描链长度作为扫描链平均值,并且利用此平均值计算出该段扫描链的离均差,再次是对每段扫描链分割出来的平均值进行第1次分配,最后将各段的所有离均差组合一起重新排列后进行第2次分配。该方法灵活性大,能结合扫描链特点进行分配,对ITC’02 SoC标准测试集上的实验结果表明本算法能得到比现有方法更平衡的结果。
IP core wrapper scan chain balance design is one of the critical methods to reduce the SoC test time. There are already many scan chain balance design algorithms, such as BFD, MVA, MVAL and so on. However,these methods all as- sign the scan chains once ,which easily sink into the dead zone of assignment. Meanwhile,in the existing methods,the MVA method uses the mean value of wrapper scan chains to predict and control the result, and can obtain more balanced results than other existing methods. However,when the internal scan chains have great dispersion,MVA method is disabled to the core and the obtained result is degraded. Aiming at these problems, the improvements are performed as follows:the scan chains are split into parts that are easy to assign,then the average of each section is used to realize the optimal reconstruc- tion of the internal scan chain again, and the balanced assignment can be realized easier. Based on above,a wrapper scan chain balance design method based on auto-section of deviation from mean (ADM) is proposed. The ADM algorithm pri- marily includes four processes. Firstly, the scan chain is segmented automatically according to the scan chain length sec- tion. Next,for each section,the scan chain length that is closest to the mean of all scan chains is taken as the average,and with this average the deviation from the mean of this scan chain is calculated. Thirdly,to the average value of each section of the scan chain, the first assignment is performed. Lastly, the deviations from the means of all the sections are combined and resorted,and then the second assignment is performed. The proposed method can flexibly assign the scan chains ac-cording to their feature. Simulation results for the embedded cores from the ITC' 02 SoC Test Benchmarks show that this al- gorithm can obtain more balanced results than other existing methods.